基于视频数据的消费者偏好测量研究

Preference measurement serves as a foundation for many marketing practice, such as new product development, pricing, advertising, etc. However, the traditional way of preference measurement techniques relies heavily on scanner data, and/or self-reported survey data. Scanner data suffers from the critic of conveying limited information. And the self-reported survey data suffers from many critics, such as involving hypothetical bias, inducing fatigue, imposing much information burden on consumers, differing from the normal decision making process, etc. In the current proposal, I propose to use video data, a new source of natural data, to supplement or even replace traditional scanner data and self-reported survey data for consumer preference measurement purpose. The motivation is twofold. The one is the wide availability of video data in various marketing contexts, such as retailing and advertising, which might be collected for other purposes like surveillance but could be potentially used to study consumer preferences. The other is superiority of video data compared to traditional scanner data and self-reported survey data. Video data is rich in information, an objective and relatively unobtrusive recording of consumer’s natural behavior, and cost efficient. Through conceptual framework and empirical study, I plan to demonstrate the effectiveness and efficiency of video data in preference measurement under advertising and retailing context. Specifically, I propose a) an intelligent video ad display system in TV advertising context; and b) a new method for ad copy testing; and c) a garment recommender model in retailing context. This research is one of the first academic research to explore the application of video data to marketing contexts, which makes contribution from both theoretical and practical perspectives. And hopefully, this research would potentially encourage fruitful future research in this area.

消费者偏好测量是众多营销实践（例如新产品开发、定价、广告等）成功的基础。但在传统营销实践中，消费者偏好测量数据大多源于扫描数据以及消费者自填的问卷数据。但是扫描数据所蕴含信息量极为有限，而消费者自填问卷数据时常不能反映消费者在购买决策时的真正偏好。因此本研究提倡使用视频数据来进行消费者偏好测量。一方面，视频数据已在营销实践中被广泛采集。例如在零售环境下，监控录像完整记录了消费者的购买决策行为，因此可作为测量消费者偏好的依据。另一方面，与传统数据相比，视频数据因其信息丰富、客观和成本低廉的特性而更显优越。通过理论构建和实验验证相结合的方式，本项目拟展示并证明视频数据在广告和零售环境中可有效测量消费者偏好。本项目拟包含三个实证子研究：1）如何构建广告智能播放系统；2）如何优化现有广告文案测评方法；3）在零售环境中如何构建服装智能推荐系统。

本项目主要研究在不同营销情境下来使用视频数据来进行消费者偏好测量的可行性和优越性。本项目共包含三个相关的子项目，目前完成的有两个，现简介如下。..第一个完成的子项目（子项目1）主要研究的是如何构建广告智能播放系统，即如何使用视频数据来有效推断消费者对于广告的偏好并据此进行实时广告推荐。研究发现，我们可以通过消费者在观看广告过程中脸部表情的变化和视线的移动来评测消费者对广告中不同视觉元素的偏好程度，并结合协同过滤算法，以此来估测消费者对于不同广告的整体偏好，从而根据不同消费者对于广告的个性化偏好选择迎合他们偏好的广告进行播放，以提高广告的沟通效果。..第二个完成的子项目（子项目3）主要研究的是在零售环境中如何构建服装智能推荐系统，即如何使用视频数据来有效推断消费者对于不同服装的喜好并据此进行实时服装推荐。研究发现，我们可以通过消费者在试穿服装过程中脸部表情的变化和手部触碰衣服位置的移动轨迹来评测消费者对于不同服装特征（例如领口，袖长，腰身等）的喜好程度，结合试穿和购买历史数据，以此来估测消费者对于服装的整体喜好程度，从而进行个性化服装推荐，以加强消费者购物体验，并提高销售。